A draft paper published in August 2026 puts a serious dent in Google’s AI Mode story. Researchers from the University of Pennsylvania and Northeastern University ran a real search experiment and found that AI Mode reduces publisher referrals while also making users less satisfied with Google Search.
What the experiment actually measured
The study enrolled 1,444 people and tracked the search behavior of 1,100 participants over seven days. A browser extension recorded how people searched, what appeared in their results, and whether they clicked through to external websites. After the experiment, 956 participants completed a survey about their experience.
Participants were randomly assigned to one of three conditions: forced regular Google Search with AI Overviews hidden, natural Google behavior, or forced AI Mode. There was one notable wrinkle: Google changed the HTML used for AI Overviews during the study, which made the no-AI condition less reliable as the week went on.
What the data showed
- AI Mode reduced click-through rates to external websites.
- When AI Overviews were removed, publisher clicks increased with no detectable decline in user experience.
- AI Mode kept users inside Google longer but sent them to websites less often.
- People assigned to AI Mode ran fewer search sessions, especially heavy Google users.
- AI Mode lowered trust, usefulness and perceived personal relevance.
- For news-related searches, there was no clear perception difference between AI Mode and regular Search.
Why this matters for performance marketers
This directly challenges Google’s claim that users love AI Mode and that AI search sends more traffic to the open web. If users are not happier with AI answers and publishers get fewer clicks, the value exchange behind SEO and performance marketing is shifting faster than many teams have planned for.
The takeaway is not to panic. It is to treat AI search visibility as a separate workstream. If your content relies on listicles or generic comparisons that AI can summarize without a visit, you need original data, proprietary examples and brand-specific detail that an AI answer cannot fully absorb.
A simple framework to respond
Run the C-A-D check on your top 20 money keywords:
- Citation: Is your brand or data showing up in AI-generated answers?
- Action: Can a user take the next step without clicking? If so, create content or tools that make a click genuinely useful.
- Diversification: Are email, YouTube, communities and direct brand sources growing fast enough to offset AI-driven referral loss?
If any part of that check comes up short, that is your next sprint. The teams that win in the AI-search era will optimize for being cited, not just for ranking.
Source: Search Engine Journal
